Walk the graph
Resolve training data → features → model → deployment from DataHub.
GET /api/sentinelNEW Deterministic lineage checks + grounded risk explanations See the sentinel ↓
ML RELIABILITY CONTROL PLANE
Catch silent model drift before users do.
Walk every DataHub lineage graph, detect upstream changes after training, and turn hidden model risk into an actionable incident brief.
✓ Deterministic checks✓ DataHub-native✓ No black-box scoring
sentinel / lineage-scanDEMO
RISK STATE1 AT RISK
82
LINEAGE TRACE
!
customer_eventsdataset · 2026-08-08
✓
feature_view_12feature · 2026-08-08
✓
churn_probability_v4model · 2026-07-18
01
Silent drift risk detectedUpstream customer events changed after the model’s last training run. Retrain against the current feature view before the next production decision.
01 / HOW IT WORKS
From graph change to clear action in one scan.
The detection path stays inspectable: timestamps establish the fact, lineage establishes impact, and the explanation makes response obvious.
Compare events
Flag upstream nodes modified strictly after the last recorded training run.
updated_at > trained_atShip context
Return evidence, severity, dates, and one concrete remediation path.
assertion: SILENT_DRIFT_RISKOperational clarity,
not another dashboard.
Evidence first
Every alert includes the exact upstream asset and change timestamp.
Safe fallback
Grounded deterministic explanations remain available without an LLM key.
Catalog-native
Write risk assertions back to DataHub where teams already investigate.